7

Let $ABC$ be a planar triangle.

Consider a random line $L$ in the plane, non parallel to any of the sides of the triangle, intersecting line $BC$ in $P$, line $AC$ in $Q$ and line $AB$ in $R$.

Question : what is the locus of points

$$G=\frac13(P+Q+R) \ \tag{0}$$

(center of weights of the 3 points) when line $L$ is varying ?

enter image description here

$$\textit{Fig. 1 : Many different points $G$ : none is inside}$$

$$\textit{the Steiner inellipse of ABC.}$$

I have an analytical answer (see proof below).

My question : is it possible to have a geometrical proof ?


My answer :

Let the barycentric coordinates of $P,Q,R$ wrt triangle $ABC$ be :

$$\begin{cases}P&(&0,&p,&(1-p)&),\\Q&(&(1-q),&0,&q&)\\R&(&r,&(1-r),&0&)\end{cases}\tag{1}$$

(where $p,q,r$ can be positive or negative).

As $P,Q,R$ are aligned, we have ;

$$\det\pmatrix{0&(1-q)&r\\p&0&(1-r)\\(1-p)&q&0}=0\tag{2}$$

which is equivalent to :

$$(1-p)(1-q)(1-r)=-pqr\tag{2'}$$

or to :

$$r=\frac{(1-p)(1-q)}{1-p-q}\tag{2''}$$

Remark : $(2),(2'),(2'')$ are different ways to express Menelaus theorem under its signed version.

Point $G$ defined by (0) has the following barycentric coordinates :

$$G \begin{cases}x&=&(1-q+r)/3,\\ y&=&(1-r+p)/3,\\z&=&(1-p+q)/3\end{cases}\tag{3}$$

(one can check that the sum of these bar. coord. is $1$).

Let us consider the following polynomial with barycentric coordinates equation (i.e., with $x+y+z=1$) :

$$f(x,y,z)=x^2+y^2+z^2-2(xy+yz+zx)\tag{3}$$

We have (see here) :

$$f(x,y,z) \begin{cases}&=&0&\text{"Steiner inellipse"} \ S_I.\\&\le&0&\text{interior of} \ S_I.\\&\ge&0&\text{exterior of} \ S_I. \end{cases}\tag{4}$$

($S_I$ is the ellipse internally tangent to the sides of $ABC$ in their midpoints).

If we "plug" the coordinates of $G$ given by (3) into the equation of $S_I$, taking into account the expression (2'') of $r$, we get :

$$f(x,y,z)=\underbrace{\frac{(p^2+q^2+pq-2p-2q+1)^2}{(1-p-q)^2}}_{ \ge 0}\tag{5}$$

As a consequence of (4), the generic point $G$ is always situated at the exterior of $S_I$.

We can conclude that the locus of points $G$ is exactly the exterior of $S_I$ because the exterior of $S_I$ is covered by ellipses $E_k$ with equations :

$$f(x,y,z)=k$$

where $k$ is any positive constant, which are homothetic images of ellipse $S_I$, the center of homothety being the centroid. For example, taking $k=1$, we obtain the outer Steiner ellipse, circumscribed to triangle $ABC$.

RobPratt
  • 50,938
Jean Marie
  • 88,997
  • I am asking myself if the above property couldn't be extended to quadrilaterals, and more generaly n-gons by using the natural generalization of Menelaus property which is Carnot's theorem. See this (far-reaching) paper. – Jean Marie Feb 02 '25 at 21:25
  • Taking the affine mapping to the equilateral triangle, where the inner ellipse maps to the incircle, might yield a simpler approach (??) $\quad$ Though even then I would just complex-geom it out with Menelaus, showing that $||a + b\omega + c\omega^2||^2 \geq 1/4$ where $abc = 1$. – Calvin Lin Feb 02 '25 at 21:41
  • @Calvin Lin Thanks for answering. Yes, good idea to take the affine mapping to triangle $1,\omega,\omega^2$. But it's hard for me to see the rationale of condition $|a+b\omega+c\omega^2|\ge 1/2$ with $abc=1$. Any reference ? [of course, such a method will still have an analytical part... but that's unimportant, if we learn a new approach]. – Jean Marie Feb 02 '25 at 22:07
  • With the centroid as the origin, and the sides of the triangle $AB, BC, CA$ corresponding to the complex numbers $1, \omega, \omega^2$, the points $P$, $Q$, $R$ would correspond to $OA + aAB = OA A+ 1$, $OB + bBC = OB + b\omega$, $OC + cCA = OC + c \omega^2$, where $abc = 1$ by Menelaus. So $ G = \frac{1}{3} {P + Q + R} = \frac{1}{3} ( OA + OB + OC + a + b\omega + c \omega^2$. We need that to fall outside of the incircle, so the norm is $ \geq 1/12$ (off by a scaling factor). – Calvin Lin Feb 03 '25 at 03:21
  • @Calvin Lin Thank you very much. Indeed an alternative approach which amounts to the minimization of $a^2+b^2+c^2-(ab+bc+ca)$ under a certain constraint but I don't agree with this constraint $abc=1$ (which isn't equivalent to Menelaus theorem) : it should be something equivalent to the relationship (1') in my question. – Jean Marie Feb 03 '25 at 08:25
  • Ah yes, should be like $2'$. So an approach would likely follow your process. – Calvin Lin Feb 03 '25 at 16:02

0 Answers0